Kubeflow Projects Releases
Kubeflow’s new release model decouples its MLOps components for faster, independent updates.
Kubeflow, the popular open-source MLOps platform under the CNCF, has announced a major structural shift. The kubeflow/kubeflow repository no longer contains the platform’s code directly. Instead, it now acts as a central hub linking to independent repositories for each project within the Kubeflow ecosystem. This includes Kubeflow Pipelines, Katib for hyperparameter tuning, the Spark Operator, Notebooks, the Model Registry, the Dashboard, and the Trainer component.
Each of these projects now follows its own release cycle, allowing teams to update or deploy individual components without waiting for a full platform release. This modular architecture mirrors the industry trend toward composable MLOps stacks, where organizations mix and match tools like MLflow, Ray, and custom solutions. The change is documented in the latest release notes, which also include a verified commit signature and redirects to the new project-specific repositories. For installation, users are directed to the official Kubeflow docs.
- Kubeflow’s monolithic repo is now a meta-repo linking to 10+ independent project repositories
- Each component (Pipelines, Katib, Spark Operator) has its own release cycle for faster updates
- The move enables teams to deploy only needed modules, reducing complexity and dependency bloat
Why It Matters
ML teams can now adopt Kubeflow components à la carte, speeding up MLOps deployments.